Memory & Cognition 1995, 23 (5), 569-580 The word frequency effect in recognition memory versus repetition priming SACHIKOKINOSHITA Macquarie University, Sydney, New South Wales, Australia The role of word frequency in recognition memory and repetition priming was investigated by using a manipulation of attention. In Experiment 1, the lexical decision task produced greater repetition priming for low-frequency words than for high-frequency words following either the attended or the unattended study condition. The recognition memory test, on the otherhand, showed a low-frequency word advantage only following the attended study condition. Furthermore, this advantage was limited to the measure of recognition memory based on conscious recollection of the study episode. In Ex­ periment 2, a speeded recognition memory test replicated the pattern obtained with the unspeeded recognition memory testin Experiment 1.These results argue against the view thatthe word frequency effects in recognition memory and repetition priming have the same origin, Instead, the results suggest that the word frequency effect in recognition memory has its locus in conscious recollection. Word frequency is a variable that has a reliable effect on enhances the feeling offamiliarity with the item. Because recognition memory tests: Low-frequency words are rec­ low-frequency items, by definition, are normally processed ognized better than high-frequency words (see, e.g., less fluently than high-frequency items, the increment in Duchek & Neely, 1989; Glanzer & Adams, 1985; Gregg, fluency due to prior exposure is proportionally greater for 1976). Word frequency also modulates the amount ofrep­ low-frequency than for high-frequency words. In this view, etition priming effects in some indirect memory tests; that the word frequency effects in repetition priming and is, facilitation due to prior exposure is greater for low-fre­ recognition memory are assumed to have their origins in a quency words than for high-frequency words under con­ common perceptual fluency component. ditions that do not make any reference to the earlier expo­ The pattern ofdata observed with amnesic Korsakoffpa­ sure. This pattern has been found in data-driven indirect tients is consistent with the preceding view. Korsakoff tests, the performance on which is assumed to be guided patients, like memory-intact subjects, show greater repe­ largely by the stimulus features; such tasks include lexical tition priming effects oflow-frequency words than ofhigh­ decision (Forster & Davis, 1984, Experiment 3; Kinoshita, frequency words in the lexical decision task (Verfaellie, 1989, Experiment 1; Scarborough, Cortese, & Scarbor­ Cermak, Letourneau, & Zuffante, 1991). In recognition ough, 1977), naming (Skinner & Grant, 1992), perceptual memory tests, Korsakoff patients have shown better per­ identification (Jacoby & Dallas, 1981), and word fragment formance with low-frequency words than with high­ and word stem completion (MacLeod, 1989; Roediger, frequency words (Huppert & Piercy, 1976; Verfaellie et al., Weldon, Stadler, & Riegler, 1992).1 1991). Because amnesic subjects perform poorly on tests Given this pattern, it seems natural to look for a common that require conscious recollection, taken together it seems basis for the word frequency effects in recognition mem­ most parsimonious to interpret these findings in terms of ory and repetition priming. Recognition memory judg­ a common perceptual fluency component intact in am­ ment is generally assumed to involve two components: a nesic subjects that underlies both repetition priming and component based on conscious recollection of the study feelings offamiliarity. episode, and another based on feelings offamiliarity (Ja­ However, recent findings reported by Gardiner and Java coby & Dallas, 1981; Mandler, 1980). A commonly held (1990), who used a procedure that specifically taps differ­ view is that prior exposure facilitates fluency ofprocess­ ent bases ofrecognition memory, cast doubt on the view ing, and that this facilitation in perceptual fluency in turn that the same mechanism is responsible for the word frequency effect found on recognition memory tests for memory-intact subjects. In this procedure, originally de­ veloped by Tulving (1985), subjects are asked to distinguish This research was supported by the Australian Research Council Grant and the Macquarie University Research Grant to S. K. Thanks are between two types of"old"responses in a recognition mem­ due Richard Castles for testing subjects. I am also grateful to the action ory test: remember responses, which are accompanied by editor, Mike Masson, and the reviewers (two anonymous, and Eyal Rein­ conscious recollections ofthe item's prior occurrence, and gold) for their helpful comments on the earlier versions ofthe manu­ know responses, which are not. A series ofstudies by Gar­ script. The stimulus materials used as well as the raw data are available diner and his associates (1988; Gardiner & Java, 1990; in the ASCII format. Correspondence should be addressed to S. Ki­ noshita, School ofBehavioural Sciences, Macquarie University, Sydney, Gardiner & Parkin, 1990) have shown dissociative effects NSW Australia 2109 (e-mail: [email protected]). of factors on the two measures. For example, Gardiner 569 Copyright 1995 Psychonomic Society, Inc. 570 KINOSHITA (1988) reported that levels-of-processing and generate­ sensitive than a direct test ofmemory (recognition mem­ versus-read-study manipulations influenced remember re­ ory test) to prior exposure when the targets were not at­ sponses but not know responses, which is consistent with tended to during study. Merikle and Reingold took this the view that these manipulations affect conceptually dri­ finding to suggest that subjects were more likely to rely on ven operations that are associated with conscious recol­ perceptual fluency than explicit recollection when words lection (cf. Roediger & Blaxton, 1987). Gardiner and Java were not consciously attended to at study. However, con­ (1990) observed that the word frequency effect was found trary to this view, Gardiner and Parkin (1990) reported that with the remember responses but not with the know re­ dividing attention at encoding reduced only the number of sponses, and hence argued that there was no evidence that remember responses and not the know responses; they thus the word frequency effect on recognition memory tests re­ argued that attention affects only conscious recollection. flects greater enhancement of perceptual fluency or fa­ There are some comments to be made against Gardiner miliarity for low- than for high-frequency words. and Parkin's (1990) claim, however. First, in their study, It may be premature, however, to rule out a role for a although the number ofknow responses was unaffected by common perceptual fluency component in the word fre­ the manipulation ofattention, the ratio ofknow responses quency effect observed in recognition memory and repe­ to overall old judgments was greater in the divided atten­ tition priming. Given that there are two bases for making tion condition than in the focused attention condition (be­ recognition memoryjudgments, a correlation between rep­ cause the number ofremember responses decreased, and etition priming and recognition memory would be reduced the number ofknow responses was unchanged). Rajaram when subjects rely on conscious recollection for making (1993) has made the point that given the nature ofinstruc­ recognition memory judgments (Whittlesea, Jacoby, & tion that requires subjects to respond know when they fail Girard, 1990). Recent studies have shown that subjects rely to make a remember response, remember and know re­ on perceptual fluency for making recognition memory sponses cannot be considered to be independent, and there­ judgments under limited conditions. For example, John­ fore the ratio, rather than the number ofresponses, may be ston, Hawley, and Elliott (1991) found a correlation be­ a more appropriate index ofwhich process subjects rely on tween the ease ofidentification ofa word and the proba­ in making recognition memoryjudgments. When this mea­ bility ofold judgments following a shallow study (vowel sure is used, Gardiner and Parkin's (1990) data are con­ counting) but not following a deep study (word naming) sistent with the view that dividing attention increases the condition. They also failed to find an effect ofexperimenter­ reliance on perceptual fluency. Second, there were more re­ induced manipulation ofperceptual fluency on the proba­ member responses than know responses even in the di­ bility ofold judgments (see also Watkins & Gibson, 1988, vided attention condition, suggesting that subjects were for a failure to find an effect ofexperimenter-induced per­ still relying predominantly on conscious recollection. Sev­ ceptual fluency). Johnston et al. concluded that perceptual eral reasons may be suggested for this. Subjects in the di­ fluency contributes to recognition memoryjudgments when vided attention condition studied each target word for it is produced naturally and when explicit memory is min­ 2 sec with intentional memory instruction while perform­ imal. Gardiner and Java (1990) used a relatively long study ing a secondary task, which consisted oflistening to a se­ duration (2 sec), and their subjects were given intentional quence oftones and classifying each tone as high, medium, memory instructions. This was likely to have produced high or low in pitch. It has been argued that